3

Shorter-Term Valuation Signals and Something About Coffee and Egg Yolks

Financial analysis relies mostly on historical data. But our goal is to make decisions that will have repercussions in the future.

—JPP

SO FAR, WE’VE FOCUSED ON RELATIVELY LONG-TERM return forecasts, or “capital markets assumptions.” These types of forecasts are useful for strategic asset allocation. Also, we saw that we can apply the equity building block model at shorter horizons, as in the equity country allocation model we discussed.

Let us now focus on shorter-term return forecasts. What are the most relevant predictive factors of returns for tactical asset allocation (TAA)? There are many macroeconomic, fundamental, and valuation signals that we can use. The key is to build a process that marries unbiased, quantitative analysis with judgment and investment experience. We must test which variables have the best forecasting power historically and put the results in the context of the current environment.

How the Tactical Process Works in Practice

To illustrate, I refer to our TAA process in the Global Multi-Asset Division at T. Rowe Price. My goal is not to advertise, but I believe in our approach. I don’t think our process is perfect (we always try to improve), and many TAA investors have been as successful as us (some more successful) with different approaches.

I prefer to focus on our process, because too many academics and practitioners explain how to forecast returns or build TAA strategies with sophisticated statistical studies and backtests, yet barely account for real-world, practical considerations. For example, the editors of the Financial Analysts Journal now reject any empirical study that doesn’t include transaction costs. The paper is not deemed worth the referees’ time—it goes straight back to the author(s) with a request to add transaction costs.

In our case, we apply this process to tactically manage asset class exposures on more than $250 billion in assets. As practitioners, we sometimes sacrifice “rigor” for simplicity and transparency. We don’t want to overfit historical data, and we obsess over whether factor models, backtests, and other useful statistical analyses are relevant given the current environment and going forward.

At its core, our TAA approach is discretionary—it’s not systematic, rules-based, or “quant.”¹ Investment judgment is key. But we rely on a wide range of quantitative inputs to frame our discussions, and our decision-making process is consistent over time. Again, I’m biased, but I believe our Asset Allocation Committee marries fundamental and quantitative investment management in a “best of both worlds” way.

There are 13 committee members from across T. Rowe’s three investment divisions: the Equity, Fixed Income, and Global Multi-Asset Divisions. [Rob Sharps, T. Rowe’s head of investments and group CIO (my boss), and I are the only two members who are not currently full-time portfolio managers, although Rob was a very successful growth equity portfolio manager before his current role. Rob cochairs the committee with Charles Shriver, portfolio manager of our Target Allocation Strategies. Charles has over 20 years of experience as a multi-asset portfolio manager and has a stellar track record.]

Our Asset Allocation Committee focuses on TAA, and nothing else. All other aspects of our business are governed by our Steering Committee, which I chair. The Steering Committee appoints members of the Asset Allocation Committee. I emphasize these details of our two-committee governance process because in my view, it’s important for investment decisions to be made independently of issues such as product design, people management, marketing, business strategy, etc. This structure means that when we discuss TAA, we don’t get bogged down with anything else that may be a waste of time for investors from other divisions. Also, it means that while I’m head of our division, I don’t own our tactical investment decisions. I’m simply a member of the TAA committee. The cochairs own our TAA decisions, so we do not vote, but it’s a collaborative process.

We meet monthly, and our time horizon is 6 to 18 months. We implement tactical changes almost every month, but rarely across all asset classes. We tend to move our portfolios incrementally, as we build overweight and underweight positions over time. The crux of our approach is to take advantage of relative valuation opportunities. We look for extreme valuation dislocations, and we like to lean against the wind—we overweight asset classes when they’re cheap, and we underweight asset classes when they’re expensive.

Importantly, we trade everything on a relative basis, and we almost always discuss asset class pairs. Our discussions often start with stocks versus bonds, or, broadly speaking, whether we want to position the portfolios “risk-on,” in which case we add to risk assets (stocks, high-yield bonds, etc.), or “risk-off,” in which case we add to more defensive asset classes (bonds, cash, etc.). Then we look for relative value opportunities at a more granular level. For example, we often ask: Are value stocks cheap relative to growth stocks? What about small versus large? Are European stocks cheap relative to US stocks? Do we like high-yield bonds relative to emerging markets bonds?

While valuation is the main driver for our decisions, we account for macro factors (growth, inflation, central bank policy, geopolitical factors, etc.), index-level fundamentals (sales, earnings, margins, leverage, etc.), and technicals (sentiment, positioning, flows, momentum, etc.). Before each meeting, every committee member pores through an up-to-date book of over 180 pages, full of all types of relevant data, return signals, and risk analytics. (I’ll discuss risk models and risk management at length in Part Two of this book.) Nonvaluation factors are used to confirm our valuation-based assessment. For example, if an asset class is cheap and other factors are positive, then our conviction is higher.

To illustrate, in early 2017, we noticed that US small caps were getting quite cheap relative to large caps. Their valuation relative to large caps was in the bottom 20%, based on data from the last 15 years and across several valuation ratios. At the same time, we thought the USD was cheap and may begin to appreciate, which tends to favor small caps over large caps; momentum was positive; M&A activity was likely to pick up in the asset class; earnings expectations were getting better; and politics were leaning toward protectionism and trade issues. It was one of those cases when the valuation, macro, fundamentals, and technical stars align. So we incrementally started to build an overweight small caps position. We continued to monitor the relevant factors, and over the next few months we built an overweight position relative to our long-term strategic allocation, which paid off.

Only in some rare cases, nonvaluation factors may drive our decision, such that we go long an asset class even though it looks expensive or short an asset class even if it looks cheap. For example, from a secular perspective (“secular”: a wonky way to say “long term” that sounds good on Bloomberg TV), we like US growth stocks, as companies such as Facebook, Apple, Netflix, and Google can make money even when the economy is sluggish. Hence, at times we overweight growth stocks even when they’re not particularly cheap relative to value stocks.

Our process illustrates how, in practice, many (perhaps most) multi-asset investors don’t forecast shorter-term returns explicitly. Rather, they evaluate a variety of factors, and they scale positions in a risk-aware framework. A return forecast doesn’t have to be a precise number with three decimals that goes into an optimizer. In our case, as I mentioned, we rely heavily on relative valuation. Over time, there’s evidence that to “buy low and sell high,” from a valuation perspective, pays off across asset classes. There’s also evidence that valuation works particularly well when momentum agrees. Hence, in addition to “buy low and sell high,” there’s something to be said for another finance bumper-sticker adage: “The trend is your friend.” (See, for example, Asness, Moskowitz, and Pedersen, 2013. We’ll discuss the combination of value and momentum in more detail in Chapter 12.)

Also, valuation tends to work better at relatively long horizons. Hence our Asset Allocation Committee’s process is slower and more incremental than those of other, more tactical investors, or “macro-gunslingers.” Our marketing department recently released a short video that shows a powerful analogy for our TAA process. In the video, an acrobat jumps about 15 feet in the air, lands on a flexible bar held by two acolytes, and rebounds again 15 feet in the air, over and over again.

A voice asks: “What makes a performance beautiful? . . . . The power lies in a series of dynamic adjustments.”

The camera zooms in on the acrobat and shows how she continually adjusts angles and balances movements and countermovements. These adjustments aren’t necessarily large.

The narrator continues: “The smallest tilt. The slightest shuffle.”

But those moves are well-timed and skillful. There’s no need for sudden and drastic moves, which could wreck the performance. This analogy represents our process well, as we often build positions over time. To take advantage of a relative valuation opportunity, we almost always move our portfolios by increments. And we’re not required to change all our positions every month. In some rare cases, we’ll even decide that we’re comfortable with our current overweight and underweight positions and not make any change.

After each meeting, I send my notes to an email distribution list of over a hundred investors throughout the firm. To give a window into our process, here is a note (“Nothing”) I sent around in late July 2018, after we decided not to make any changes to our positions. Notice how we rely on a wide range of signals, how we incorporate our views into the process, and how we focus on relative valuations. For our TAA decisions to be well-timed (to deliver a beautiful performance), we need a lot of building blocks for our process, each of which must be additive to the others.

Which Valuation Signal Works Best for Tactical Asset Allocation?

In 2018, I tested the effectiveness of three valuation metrics as predictors of stock returns: price to earnings (P/E), price to book (P/B), and price to cash flow (P/CF). This test was similar to the project I mentioned in Chapter 2, when I evaluated which ratio best predicted valuation changes. To make it simpler, this time I focused on total returns. I also expanded the analysis to various time horizons and compared the effectiveness of valuation ratios as predictors of absolute and relative performance between asset class pairs.

I used the equity asset classes, shown in Table 3.1, that we’ve used so far in this book when we discussed the CAPM and other long-term return forecasts. I estimated the relationship between initial valuation levels and subsequent returns at the six-month, one-year, two-year, and three-year horizons for 10 equity asset classes and 6 relative bets between them. My monthly data sample started in January 1995 and ended in May 2018.² In total, I ran 192 out-of-sample correlations between signal and forward returns.

TABLE 3.1 List of Equity Asset Classes and Relative Bets

Normally I would code such a repetitive empirical analysis in Matlab (a technical engineering and finance software program) and use Microsoft Excel to visualize the data and test pieces of my code. But I’m an inefficient and out-of-practice programmer, so I used Bloomberg as my source of historical data and ran everything in Excel. It made for a fun afternoon.

Some of the results surprised me. Valuation seemed to work consistently across equity asset classes and time horizons. Out of 192 correlations, across absolute and relative return forecasts, 187 had the expected sign, consistent with the “buy cheap (valuations), sell high (valuations)” intuition—a remarkable hit rate of 97%. And some of the failed signals seemed to be driven by outlier P/E data around the tech bubble. Also, it was surprising that the signals seemed to work well at shorter horizons (six months and one year).

But I should emphasize that there’s a difference between a result that shows the expected sign on a correlation and the ability to make money with the signal. Average correlations were in the –20% range for six-month and one-year horizons and strengthened to –30 to –40% for two- and three-year horizons. (Valuation tends to work better at longer time horizons.) These numbers aren’t very high: –40% corresponds to an R-square of 16%. (I used a convention whereby the expected sign on the correlation is negative, because high valuations lead to lower returns, and vice versa. In that context, the more negative the correlation, the more predictive the signal.) Plenty of times, the valuation signal will lead you astray. Over the last few years, for example, non-US equities have been consistently cheaper than US equities, but US markets have continued to outperform.

Consistent with my prior valuation horse race, to forecast absolute returns, P/CF worked better than P/B and P/E. At the one-year horizon, P/CF had an average correlation of over –43% with forward returns, compared with –26% for P/B and –22% for P/E.

However, when I ran the tests for relative returns, for example, large cap versus small cap stocks, the results surprised me. P/E worked better than P/CF at the six-month, one-year, and two-year horizons and worked about the same as P/CF at the three-year horizon. Why didn’t P/CF win? What happened to my simple explanation that cash flows are harder to game and thereby more reliable?

When I struggle with a tricky question like this one, I usually walk out of my office, turn left, and find someone from our research team to help me think through the problem. As long as one member of this talented team is available to indulge my random questions, I know I’ll get thoughtful answers, good hypotheses, and help thinking through the problem. Not every investor has access to such a team. Nonetheless, when forecasting returns, I believe it’s important to always run your assumptions by someone, even nonexperts, because verbalizing your thoughts to someone often leads to breakthroughs.

This time Chris Faulkner-MacDonagh, who has a Yale doctorate and an impressive résumé, was happy to talk about my question: “Why would an indicator work well to forecast time series returns for individual assets, but not work well to forecast relative returns between those same assets?”

Off the cuff, he suggested two explanations. First, the obvious: “It might be an issue with autocorrelation. A variable may seem to work well as a time series forecast if it’s highly autocorrelated (persistent), even if it’s not particularly predictive. So if cash flows are more autocorrelated than other variables, perhaps your result for P/CF was spurious.”

In other words, he suggested P/CF “cheated.” He posited that cash flows may be more autocorrelated than earnings. However, a quick test revealed that this conclusion didn’t apply to my asset classes. Chris was probably on to something, and there’s a whole field of study in econometrics dedicated to these types of issues with autocorrelation, unit roots, cointegration, etc. But I was after the big picture, so I stepped away from this methodological rabbit hole.

His second suggestion was thoughtful as well: “If you use index-adjusted positive earnings, remember that negative earnings are removed from the aggregation, which makes asset class–level earnings less outlier-sensitive and more comparable across asset classes.”

Good point. Adjusted positive earnings may be more comparable across asset classes, even though they’re less predictive from a time series perspective, for a given asset class.

I also went to talk to Rob Panariello, portfolio manager and quantitative analyst. He’s our expert on portfolio optimization and risk models. (Also, unlike many people with such an incredible intellect, Rob has a fantastic sense of humor and doesn’t take himself too seriously. It’s outside the scope of this book, but here are a few fun facts about Rob: he wears slippers to work; he grows strange plants in his office; he drinks 7–10 coffees every day; he used to be a power lifter; he regularly wins competitive taco-eating contests; and he owns an isolation tank where he likes to relax for hours at a time.)

I’ve coauthored papers with Rob. He has one of the most intuitive minds I’ve ever come across. He grasps issues quickly. When I asked him my question, he nailed the answer in two words: “Systemwide noise.”

Of course! The time series forecasts may be completely off for two asset classes, but if the bias is systemwide—if it affects both asset classes the same way—then the comparison between the two can still be meaningful. It appears P/Es are more subject to systemwide noise than are P/CFs. So they perform better for relative return (“cross-sectional”) forecasts than for time series forecasts.

In general, time series forecasts are useful for market timing. For overweight and underweight decisions, cross-sectional forecasts are what matters. Good cross-sectional forecasts can be poor time series forecasts, and vice versa. The reason? Systemwide noise. Another way to think about this important distinction is that some investors may seek to predict when to invest in stocks (a time series question), while others may focus on whether value stocks will outperform growth stocks (a cross-sectional question).

Ultimately, whether the decision is about market timing or relative valuation (which is our focus in our own process), I don’t think investors should rely on one signal, whether it’s P/CF, P/E, or something else. In our tactical asset allocation process we use a variety of valuation ratios. We look for which ratios have been most predictive for each asset class pair, and we focus on them. We also build composite metrics across several indicators. Importantly, we always evaluate the data in the context of the current environment.

Relative Valuation Between Stocks and Bonds and Across Bond Markets

In Chapter 2, I mentioned a study by my colleagues Justin Harvey and Aaron Stonacek that showed that yield to maturity provides a simple and remarkably effective forecast for bond returns, especially when the investment horizon is close to the duration of the portfolio. What about relative returns between stocks and bonds and across fixed income asset classes?

For US stocks versus bonds, our research team’s analysis shows a 35% correlation between the ratio of forward equity earnings yield (the inverse of the P/E ratio) to bond yield and the subsequent 12-month relative returns, based on monthly data from January 1990 to June 2018.³ The intuition behind this signal is straightforward: when bond yield increases, bonds become cheaper; when earnings yield increases (P/E decreases), stocks become cheaper. So we keep track of the ratio between the two asset classes and try to “buy low and sell high.” It’s a decent signal, but it’s far from perfect. It’s mostly useful from a big-picture perspective. For example, rates have been so low in recent years that many investors have stayed in stocks despite high P/E ratios. We’ve had these discussions several times in our Asset Allocation Committee. One member would say: “Stocks are expensive. We’ve rarely seen P/Es this high.”

Then another member would respond: “Yes, but have you looked at rates? Bonds are even more expensive.”

Such is the plight of the tactical asset allocator. The important takeaway is that a P/E of say, 16, means different things if the 10-year yield is at 2% or 4%. One could argue that stocks are more “expensive” when the 10-year yield is at 4%, because they’re less attractive relative to bonds—in other words, the equity risk premium is compressed. In an award-winning paper,⁴ Arnott, Chaves, and Chow (2017) show that the Shiller P/E performs better as a short-term timing signal if we adjust it for inflation and real rates.

Yet we shouldn’t push this intuition too far. The authors explain that the relationship is not linear. There’s a goldilocks level of inflation and real rates that appear “just right” to justify high valuations. Ultimately, if the prediction horizon is one year, a 35% correlation indicates that many other factors matter: momentum, expected earnings, monetary policy, sentiment, etc. And the related, popular narrative that rising rates are bad for stocks can be misleading. The Fed often raises rates when growth accelerates and risk assets perform well. In fact, when rates have increased by 25 bps or more between January 1990 and February 2018, average stock returns have been positive.⁵ At an Asset Allocation Committee meeting in April 2018, I titled my presentation, “What Do Rising Rates Have in Common with Coffee and Egg Yolks?” And I concluded with the following analogy:

It’s never been clear whether coffee and egg yolks are good or bad for your health, but everybody seems to have a black-or-white opinion on either side of the debate. The reality is probably more complex due to a variety of factors. Are rising rates bad for stocks? I don’t think we should make such bold statements in the current environment. Rather, we should worry about softening economic growth and high valuations.

[Charles Shriver, cochair of the committee, then suggested that the same inconclusiveness applies to red wine: Is it good or bad for your health? In the same vein, at the end of my presentation, I added a reference to a study on running, which shows that running is good for your knees—clearly a controversial conclusion. (I’m diversified across these risks: I love red wine, I’m an avid coffee drinker, I eat a hard-boiled egg every day with my lunch salad, and I run about 20–30 miles per week.)]

Our chief US economist, Alan Levenson, offered a thoughtful response to my presentation. He provided some intuition for the relatively low power of the stocks versus bonds, yield-based valuation signal:

The Fed’s efforts to either cap inflation or stop the unemployment rate from falling eventually contribute to recessions. But I agree that in the early stages of Fed rate hikes, risk assets can rise alongside risk-free rates, because monetary policy is still simulative (right now, the real fed funds rate is roughly zero). If growth data stabilize/improve in the near term (and barring political/geopolitical disruption), rates and risk assets can rise together. The challenge is when policy becomes tighter (say, 100 bps from now on real fed funds) and economic fundamentals begin to roll over.

Moreover, it’s not always clear whether the Fed leads the economy, or vice versa. Stefan Hubrich, portfolio manager, put it this way:

Mistaken causality? The naïve interpretation is that rising rates cause the recession. Is it possible that instead, the oncoming recession caused the Fed to lower rates? “Recessions end tightening cycles” rather than “tightening cycles cause recessions”?

Therefore, the yield ratio for stocks versus bonds should be used with caution, and in the context of other factors. It’s one piece of the puzzle.

In contrast, yield ratios across fixed income asset classes work quite well as stand-alone signals. In fact, some of these ratios may be the most predictive TAA signals we have in our arsenal. For example, the yield ratio between emerging markets bonds and US investment-grade bonds has a 70% correlation with 12-month forward relative returns; for US high yield versus US investment-grade bonds, the correlation is 61%; and for US high yield versus emerging markets bonds, the correlation is 50%.⁶

This level of predictability doesn’t seem attainable in equity markets. In the series of tests on P/E, P/B, and P/CF we discussed earlier, the average correlation across equity relative signals was –22% (a reminder that the sign was reversed simply by convention). Across 24 signals, only 2 had correlations stronger than –40%: P/E as a signal for US value versus growth stocks (–48%) and P/E as a signal for world ex-US versus US equities (–43%). The lesson is that relative valuation signals work better in fixed income markets than in equity markets.

There might be ways to improve on simple yield ratio signals. In the monograph I coauthored with my former PIMCO colleagues Vasant Naik, Mukundan Devarajan, Andrew Nowobilski, and Niels Pedersen (2016), titled “Factor Investing and Asset Allocation: A Business Cycle Perspective,” we use a definition of carry that includes roll down, based on the difference between spot and 12-month forward swap rates. My coauthors simulated a tactical strategy that ranks carry across six countries: United States, Germany, Japan, the United Kingdom, Australia, and Canada. The strategy goes long the top three carry markets and short the bottom three. It generates a Sharpe ratio of 0.73 between 2002 and 2015 (based on quarterly data).

Notes

1.   Although our TAA process is broadly applied across almost all our portfolios, I don’t want to discount systematic strategies, such as managed volatility, covered call writing, and dynamic risk premiums, which we also use in several portfolios and in our custom solutions. Such strategies have some advantages that discretionary approaches don’t have—for example, they eliminate behavioral biases. I’ll discuss these strategies in Part Two of this book. Also, one of our portfolio managers, Rick de los Reyes, runs a discretionary “best-ideas” portfolio that incorporates different multi-asset tactical approaches.

2.   Bloomberg Finance L.P. Monthly data from January 31, 1995 to May 31, 2018. Data fields used are TOT_RETURN_INDEX_GROSS_DVDS, INDX_ADJ_POSITIVE_PE, PX_TO_BOOK_RATIO, and PX_TO_CASH_FLOW. Backtests involving EAFE small start on January 31, 1998, as data are not available going back to 1995.

3.   Monthly data from January 1990 to June 2018, Russell 3000 Fwd. Earnings Yield versus Real YTW (YTW less Y/Y Core CPI) for the Bloomberg Barclays Aggregate. Sources are T. Rowe Price, Bloomberg Finance L.P., and BLS. Analysis by Chris Faulkner-MacDonagh and David Clewell. Unlike in the previous section, here I quote predictive correlations as a positive number, because the signal relies on yields. A high yield, ceteris paribus, means an asset class is cheap. For equities, yield is simply the inverse of the P/E ratio.

4.   The article won the 19th Annual Bernstein Fabozzi/Jacobs Levy Award for best article appearing in the Journal of Portfolio Management during 2017.

5.   Using monthly data from January 1990 to February 2018 on the Russell 3000.

6.   Barclays US Aggregate for US investment grade (Source: Bloomberg Finance L.P., yield to worst and total return, time period based on other data series availabilities); JP Morgan EMBI Global for Emerging Market Bonds (Source: JP Morgan, yield to worst and total return from December 2001 to June 2018); JP Morgan U.S. High Yield (Source: JP Morgan, yield to worst and total return from January 1999 to June 2018; start in December 2001 for correlation with EMBI).